Performance improvements (#337, #338, #339, #341)
eval_select() out-of-bounds errors now use the verb “select” rather than “subset” in the error message for consistency with dplyr::select() (#271).
Fix for CRAN checks.
New tidyselect_data_proxy() and tidyselect_data_has_predicates() allows tidyselect to work with custom input types (#242).
New eval_relocate() for moving a selection. This powers dplyr::relocate() (#232).
Using all_of() outside of a tidyselect context is now deprecated (#269). In the future it will error to be consistent with any_of().
Use of .data in tidyselect expressions is now deprecated to more cleanly separate tidy-select from data-masking. Replace .data$x with "x" and .data[[var]] with all_of(var) (#169).
Use of bare predicates (not wrapped in where()) and indirection (without using all_of()) have been formally deprecated (#317).
Selection language:
any_of() generates a more informative error if you supply too many arguments (#241).
all_of() (like any_of()) returns an integer vector to make it easier to combine in functions (#270, #294). It also fails when it can’t find variables even when strict = FALSE.
matches() recognises and correctly uses stringr pattern objects (stringr::regex(), stringr::fixed(), etc) (#238). It also now works with named vectors (#250).
num_range() gains a suffix argument (#229).
where() is now exported, like all other select helpers (#201), and gives more informative errors (#236).
eval_select() with include now preserves the order of the variables if they’re present in the selection (#224).
eval_select() always returns a named vector, even when renaming is not permitted (#220).
eval_select() and eval_relocate() gain new allow_empty argument which makes it possible to forbid empty selections with allow_empty = FALSE (#252).
eval_select(allow_rename = FALSE) no longer fails with empty selections (#221, @eutwt) or with predicate functions (#225). It now properly fails with partial renaming (#305).
peek_var() error now generates hyperlink to docs with recent RStudio (#289).
vars_pull() generates more informative error messages (#234, #258, #318) and gains error_call and error_arg arguments.
Errors produced by tidyselect should now be more informative. Evaluation errors are now chained, with the child error call is set to the error_call argument of eval_select() and eval_rename(). We’ve also improved backtraces of base errors, and done better at propagating the root error_call to vctrs input checkers.
tidyselect_verbosity is no longer used; deprecation messaging is now controlled by lifecycle_verbosity like all other packages (#317).
Fix for CRAN checks.
Better compatibility with rlang 1.0.0 errors. More to come soon.
Fix for CRAN checks.
tidyselect has been re-licensed as MIT (#217).
Predicate functions must now be wrapped with where().
iris %>% select(where(is.factor))We made this change to avoid puzzling error messages when a variable is unexpectedly missing from the data frame and there is a corresponding function in the environment:
# Attempts to invoke `data()` function
data.frame(x = 1) %>% select(data)Now tidyselect will correctly complain about a missing variable rather than trying to invoke a function.
For compatibility we will support predicate functions starting with is for 1 version.
eval_select() gains an allow_rename argument. If set to FALSE, renaming variables with the c(foo = bar) syntax is an error. This is useful to implement purely selective behaviour (#178).
Fixed issue preventing repeated deprecation messages when tidyselect_verbosity is set to "verbose" (#184).
any_of() now preserves the order of the input variables (#186).
The return value of eval_select() is now always named, even when inputs are constant (#173).
This is the 1.0.0 release of tidyselect. It features a more solidly defined and implemented syntax, support for predicate functions, new boolean operators, and much more.
New Get started vignette for client packages. Read it with vignette("tidyselect") or at https://tidyselect.r-lib.org/articles/tidyselect.html.
The definition of the tidyselect language has been consolidated. A technical description is now available: https://tidyselect.r-lib.org/articles/syntax.html.
all_of() instead. Referring to contextual objects with a bare name is brittle because it might be masked by a data frame column. Using all_of() is safe (#76).tidyselect now uses vctrs for validating inputs. These changes may reveal programming errors that were previously silent. They may also cause failures if your unit tests make faulty assumptions about the content of error messages created in tidyselect:
Out-of-bounds errors are thrown when a name doesn’t exist or a location is too large for the input.
Logical vectors now fail properly.
Selected variables now must be unique. It was previously possible to return duplicate selections in some circumstances.
The input names can no longer contain NA values.
Note that we recommend testthat::verify_output() for monitoring error messages thrown from packages that you don’t control. Unlike expect_error(), verify_output() does not cause CMD check failures when error messages have changed. See https://www.tidyverse.org/blog/2019/11/testthat-2-3-0/ for more information.
The boolean operators can now be used to create selections (#106).
! negates a selection.| takes the union of two selections.& takes the intersection of two selections.These patterns can currently be achieved using -, c() and intersect() respectively. The boolean operators should be more intuitive to use.
Many thanks to Irene Steves (@isteves) for suggesting this UI.
You can now use predicate functions in selection contexts:
This feature is not available in functions that use the legacy interface of tidyselect. These need to be updated to use the new eval_select() function instead of vars_select().
Unary - inside nested c() is now consistently syntax for set difference (#130).
Improved support for named elements. It is now possible to assign the same name to multiple elements, if the input data structure doesn’t require unique names (i.e. anything but a data frame).
The selection engine has been rewritten to support a clearer separation between data-expressions (calls to :, -, and c) and env-expressions (anything else). This means you can now safely use expressions of the type:
Even if the data frame data contains a column also named data, the subexpression ncol(data) is still correctly evaluated. The data:ncol(data) expression is equivalent to 2:3 because data is looked up in the relevant context without ambiguity:
data <- tibble(foo = 1, data = 2, bar = 3)
data %>% dplyr::select(data:ncol(data))
#> # A tibble: 1 x 2
#> data bar
#> <dbl> <dbl>
#> 1 2 3While this example above is a bit contrived, there are many realistic cases where these changes make it easier to write safe code:
select_from <- function(data, var) {
data %>% dplyr::select({{ var }} : ncol(data))
}
data %>% select_from(data)
#> # A tibble: 1 x 2
#> data bar
#> <dbl> <dbl>
#> 1 2 3The new selection helpers all_of() and any_of() are strict variants of one_of(). The former always fails if some variables are unknown, while the latter does not. all_of() is safer to use when you expect all selected variables to exist. any_of() is useful in other cases, for instance to ensure variables are selected out:
vars <- c("Species", "Genus")
iris %>% dplyr::select(-any_of(vars))Note that all_of() and any_of() are a bit more conservative in their function signature than one_of(): they do not accept dots. The equivalent of one_of("a", "b") is all_of(c("a", "b")).
Selection helpers like all_of() and starts_with() are now available in all selection contexts, even when they haven’t been attached to the search path. The most visible consequence of this change is that it is now easier to use selection functions without attaching the host package:
# Before
dplyr::select(mtcars, dplyr::starts_with("c"))
# After
dplyr::select(mtcars, starts_with("c"))It is still recommended to export the helpers from your package so that users can easily look up the documentation with ?.
starts_with(), ends_with(), contains(), and matches() now accept vector inputs (#50). For instance these are now equivalent ways of selecting all variables that start with either "a" or "b":
starts_with(c("a", "b"))
starts_with("a") | starts_with("b")matches() has new argument perl to allow for Perl-like regular expressions (@fmichonneau, #71)
Better support for selecting with S3 vectors. For instance, factors are treated as characters.
New eval_select() and eval_rename() functions for client packages. These replace vars_select() and vars_rename(), which are now deprecated. These functions:
Take the full data rather than just names. This makes it possible to use function predicates in selection context.
Return a numeric vector of locations rather than a vector of names. This makes it possible to use tidyselect with inputs that support duplicate names, like regular vectors.
The .strict argument of vars_select() now works more robustly and consistently.
Using arithmetic operators in selection context now fails more informatively (#84).
It is now possible to select columns in data frames containing duplicate variables (#94). However, the duplicates can’t be part of the final selection.
eval_rename() no longer ignore the names of unquoted character vectors of length 1 (#79).
eval_rename() now fails when a variable is renamed to an existing name (#70).
eval_rename() has better support for existing duplicates (but creating new duplicates is an error).
eval_select(), eval_rename() and vars_pull() now detect missing values uniformly (#72).
vars_pull() now includes the faulty expression in error messages.
The performance issues of eval_rename() with many arguments have been fixed. This make dplyr::rename_all() with many columns much faster (@zkamvar, #92).
tidyselect is now much faster with many columns, thanks to a performance fix in rlang::env_bind() as well as internal fixes.
vars_select() ignores vectors with only zeros (#82).
This is a maintenance release for compatibility with rlang 0.3.0.
Fixed a warning that occurred when a vector of column positions was supplied to vars_select() or functions depending on it such as tidyr::gather() (#43 and tidyverse/tidyr#374).
Fixed compatibility issue with rlang 0.2.0 (#51).
Internal fixes in prevision of using tidyselect within dplyr.
vars_select() and vars_rename() now correctly support unquoting character vectors that have names.
vars_select() now ignores missing variables.
dplyr is now correctly mentioned as suggested package.- now supports character vectors in addition to strings. This makes it easy to unquote column names to exclude from the set:
vars <- c("cyl", "am", "disp", "drat")
vars_select(names(mtcars), - !!vars)last_col() now issues an error when the variable vector is empty.
last_col() now returns column positions rather than column names for consistency with other helpers. This also makes it compatible with functions like seq().
c() now supports character vectors the same way as - and seq(). (#37 @gergness)
The main point of this release is to revert a troublesome behaviour introduced in tidyselect 0.1.0. It also includes a few features.
The special evaluation semantics for selection have been changed back to the old behaviour because the new rules were causing too much trouble and confusion. From now on data expressions (symbols and calls to : and c()) can refer to both registered variables and to objects from the context.
However the semantics for context expressions (any calls other than to : and c()) remain the same. Those expressions are evaluated in the context only and cannot refer to registered variables.
If you’re writing functions and refer to contextual objects, it is still a good idea to avoid data expressions. Since registered variables are change as a function of user input and you never know if your local objects might be shadowed by a variable. Consider:
n <- 2
vars_select(letters, 1:n)Should that select up to the second element of letters or up to the 14th? Since the variables have precedence in a data expression, this will select the 14 first letters. This can be made more robust by turning the data expression into a context expression:
vars_select(letters, seq(1, n))You can also use quasiquotation since unquoted arguments are guaranteed to be evaluated without any user data in scope. While equivalent because of the special rules for context expressions, this may be clearer to the reader accustomed to tidy eval:
vars_select(letters, seq(1, !! n))Finally, you may want to be more explicit in the opposite direction. If you expect a variable to be found in the data but not in the context, you can use the .data pronoun:
vars_select(names(mtcars), .data$cyl : .data$drat)The new select helper last_col() is helpful to select over a custom range: vars_select(vars, 3:last_col()).
: and - now handle strings as well. This makes it easy to unquote a column name: (!!name) : last_col() or - !!name.
vars_select() gains a .strict argument similar to rename_vars(). If set to FALSE, errors about unknown variables are ignored.
vars_select() now treats NULL as empty inputs. This follows a trend in the tidyverse tools.
vars_rename() now handles variable positions (integers or round doubles) just like vars_select() (#20).
vars_rename() is now implemented with the tidy eval framework. Like vars_select(), expressions are evaluated without any user data in scope. In addition a variable context is now established so you can write rename helpers. Those should return a single round number or a string (variable position or variable name).
has_vars() is a predicate that tests whether a variable context has been set (#21).
The selection helpers are now exported in a list vars_select_helpers. This is intended for APIs that embed the helpers in the evaluation environment.
one_of() argument vars has been renamed to .vars to avoid spurious matching.tidyselect is the new home for the legacy functions dplyr::select_vars(), dplyr::rename_vars() and dplyr::select_var().
We took this opportunity to make a few changes to the API:
select_vars() and rename_vars() are now vars_select() and vars_rename(). This follows the tidyverse convention that a prefix corresponds to the input type while suffixes indicate the output type. Similarly, select_var() is now vars_pull().
The arguments are now prefixed with dots to limit argument matching issues. While the dots help, it is still a good idea to splice a list of captured quosures to make sure dotted arguments are never matched to vars_select()’s named arguments:
vars_select(vars, !!! quos(...))Error messages can now be customised. For consistency with dplyr, error messages refer to “columns” by default. This assumes that the variables being selected come from a data frame. If this is not appropriate for your DSL, you can now add an attribute vars_type to the .vars vector to specify alternative names. This must be a character vector of length 2 whose first component is the singular form and the second is the plural. For example, c("variable", "variables").
tidyselect provides a few more ways of establishing a variable context:
scoped_vars() sets up a variable context along with an an exit hook that automatically restores the previous variables. It is the preferred way of changing the variable context.
with_vars() takes variables and an expression and evaluates the latter in the context of the former.
poke_vars() establishes a new variable context. It returns the previous context invisibly and it is your responsibility to restore it after you are done. This is for expert use only.
current_vars() has been renamed to peek_vars(). This naming is a reference to peek and poke from legacy languages.
The evaluation semantics for selecting verbs have changed. Symbols are now evaluated in a data-only context that is isolated from the calling environment. This means that you can no longer refer to local variables unless you are explicitly unquoting these variables with !!, which is mostly for expert use.
Note that since dplyr 0.7, helper calls (like starts_with()) obey the opposite behaviour and are evaluated in the calling context isolated from the data context. To sum up, symbols can only refer to data frame objects, while helpers can only refer to contextual objects. This differs from usual R evaluation semantics where both the data and the calling environment are in scope (with the former prevailing over the latter).